Late Fusion

Late fusion is a data fusion technique that combines predictions or features from multiple independent models, rather than integrating data at earlier processing stages. Current research focuses on optimizing late fusion strategies across diverse applications, including object detection, image segmentation, and video classification, often employing neural networks like UNets and Vision Transformers, or simpler methods like averaging predictions or using Support Vector Machines. This approach offers advantages in computational efficiency and robustness to noisy or incomplete data from individual modalities, leading to improved accuracy and reliability in various fields such as autonomous driving, medical image analysis, and multimedia content analysis.